English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Point cloud classifiers with rotation robustness have been widely discussedin the 3D deep learning community. Most proposed methods either use rotationinvariant descriptors as inputs or try to design rotation equivariant networks.However, robust models generated by these methods have limited performanceunder clean aligned datasets due to modifications on the original classifiersor input space. In this study, for the first time, we show that the rotationrobustness of point cloud classifiers can also be acquired via adversarialtraining with better performance on both rotated and clean datasets.Specifically, our proposed framework named ART-Point regards the rotation ofthe point cloud as an attack and improves rotation robustness by training theclassifier on inputs with Adversarial RoTations. We contribute an axis-wiserotation attack that uses back-propagated gradients of the pre-trained model toeffectively find the adversarial rotations. To avoid model over-fitting onadversarial inputs, we construct rotation pools that leverage thetransferability of adversarial rotations among samples to increase thediversity of training data. Moreover, we propose a fast one-step optimizationto efficiently reach the final robust model. Experiments show that our proposedrotation attack achieves a high success rate and ART-Point can be used on mostexisting classifiers to improve the rotation robustness while showing betterperformance on clean datasets than state-of-the-art methods.

ART-Point: Improving Rotation Robustness of Point Cloud Classifiers via Adversarial Rotation